git.ozlabs.org Git - ccan/blob - ccan/tdb2/tdb.c

   1 #include "private.h"
   2 #include <ccan/tdb2/tdb2.h>
   3 #include <assert.h>
   4 #include <stdarg.h>
   5
   6 /* The null return. */
   7 struct tdb_data tdb_null = { .dptr = NULL, .dsize = 0 };
   8
   9 /* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
  10 static struct tdb_context *tdbs = NULL;
  11
  12 static bool tdb_already_open(dev_t device, ino_t ino)
  13 {
  14         struct tdb_context *i;
  15
  16         for (i = tdbs; i; i = i->next) {
  17                 if (i->device == device && i->inode == ino) {
  18                         return true;
  19                 }
  20         }
  21
  22         return false;
  23 }
  24
  25 static bool read_all(int fd, void *buf, size_t len)
  26 {
  27         while (len) {
  28                 ssize_t ret;
  29                 ret = read(fd, buf, len);
  30                 if (ret < 0)
  31                         return false;
  32                 if (ret == 0) {
  33                         /* ETOOSHORT? */
  34                         errno = EWOULDBLOCK;
  35                         return false;
  36                 }
  37                 buf = (char *)buf + ret;
  38                 len -= ret;
  39         }
  40         return true;
  41 }
  42
  43 static uint64_t random_number(struct tdb_context *tdb)
  44 {
  45         int fd;
  46         uint64_t ret = 0;
  47         struct timeval now;
  48
  49         fd = open("/dev/urandom", O_RDONLY);
  50         if (fd >= 0) {
  51                 if (read_all(fd, &ret, sizeof(ret))) {
  52                         tdb_logerr(tdb, TDB_SUCCESS, TDB_DEBUG_TRACE,
  53                                  "tdb_open: random from /dev/urandom");
  54                         close(fd);
  55                         return ret;
  56                 }
  57                 close(fd);
  58         }
  59         /* FIXME: Untested!  Based on Wikipedia protocol description! */
  60         fd = open("/dev/egd-pool", O_RDWR);
  61         if (fd >= 0) {
  62                 /* Command is 1, next byte is size we want to read. */
  63                 char cmd[2] = { 1, sizeof(uint64_t) };
  64                 if (write(fd, cmd, sizeof(cmd)) == sizeof(cmd)) {
  65                         char reply[1 + sizeof(uint64_t)];
  66                         int r = read(fd, reply, sizeof(reply));
  67                         if (r > 1) {
  68                                 tdb_logerr(tdb, TDB_SUCCESS, TDB_DEBUG_TRACE,
  69                                            "tdb_open: %u random bytes from"
  70                                            " /dev/egd-pool", r-1);
  71                                 /* Copy at least some bytes. */
  72                                 memcpy(&ret, reply+1, r - 1);
  73                                 if (reply[0] == sizeof(uint64_t)
  74                                     && r == sizeof(reply)) {
  75                                         close(fd);
  76                                         return ret;
  77                                 }
  78                         }
  79                 }
  80                 close(fd);
  81         }
  82
  83         /* Fallback: pid and time. */
  84         gettimeofday(&now, NULL);
  85         ret = getpid() * 100132289ULL + now.tv_sec * 1000000ULL + now.tv_usec;
  86         tdb_logerr(tdb, TDB_SUCCESS, TDB_DEBUG_TRACE,
  87                    "tdb_open: random from getpid and time");
  88         return ret;
  89 }
  90
  91 struct new_database {
  92         struct tdb_header hdr;
  93         struct tdb_freetable ftable;
  94 };
  95
  96 /* initialise a new database */
  97 static int tdb_new_database(struct tdb_context *tdb,
  98                             struct tdb_attribute_seed *seed,
  99                             struct tdb_header *hdr)
 100 {
 101         /* We make it up in memory, then write it out if not internal */
 102         struct new_database newdb;
 103         unsigned int magic_len;
 104         ssize_t rlen;
 105
 106         /* Fill in the header */
 107         newdb.hdr.version = TDB_VERSION;
 108         if (seed)
 109                 newdb.hdr.hash_seed = seed->seed;
 110         else
 111                 newdb.hdr.hash_seed = random_number(tdb);
 112         newdb.hdr.hash_test = TDB_HASH_MAGIC;
 113         newdb.hdr.hash_test = tdb->khash(&newdb.hdr.hash_test,
 114                                          sizeof(newdb.hdr.hash_test),
 115                                          newdb.hdr.hash_seed,
 116                                          tdb->hash_priv);
 117         newdb.hdr.recovery = 0;
 118         memset(newdb.hdr.reserved, 0, sizeof(newdb.hdr.reserved));
 119         /* Initial hashes are empty. */
 120         memset(newdb.hdr.hashtable, 0, sizeof(newdb.hdr.hashtable));
 121
 122         /* Free is empty. */
 123         newdb.hdr.free_table = offsetof(struct new_database, ftable);
 124         memset(&newdb.ftable, 0, sizeof(newdb.ftable));
 125         set_header(NULL, &newdb.ftable.hdr, TDB_FTABLE_MAGIC, 0,
 126                    sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
 127                    sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr), 0);
 128
 129         /* Magic food */
 130         memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
 131         strcpy(newdb.hdr.magic_food, TDB_MAGIC_FOOD);
 132
 133         /* This creates an endian-converted database, as if read from disk */
 134         magic_len = sizeof(newdb.hdr.magic_food);
 135         tdb_convert(tdb,
 136                     (char *)&newdb.hdr + magic_len, sizeof(newdb) - magic_len);
 137
 138         *hdr = newdb.hdr;
 139
 140         if (tdb->flags & TDB_INTERNAL) {
 141                 tdb->map_size = sizeof(newdb);
 142                 tdb->map_ptr = malloc(tdb->map_size);
 143                 if (!tdb->map_ptr) {
 144                         tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
 145                                    "tdb_new_database: failed to allocate");
 146                         return -1;
 147                 }
 148                 memcpy(tdb->map_ptr, &newdb, tdb->map_size);
 149                 return 0;
 150         }
 151         if (lseek(tdb->fd, 0, SEEK_SET) == -1)
 152                 return -1;
 153
 154         if (ftruncate(tdb->fd, 0) == -1)
 155                 return -1;
 156
 157         rlen = write(tdb->fd, &newdb, sizeof(newdb));
 158         if (rlen != sizeof(newdb)) {
 159                 if (rlen >= 0)
 160                         errno = ENOSPC;
 161                 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_FATAL,
 162                            "tdb_new_database: %zi writing header: %s",
 163                            rlen, strerror(errno));
 164                 return -1;
 165         }
 166         return 0;
 167 }
 168
 169 struct tdb_context *tdb_open(const char *name, int tdb_flags,
 170                              int open_flags, mode_t mode,
 171                              union tdb_attribute *attr)
 172 {
 173         struct tdb_context *tdb;
 174         struct stat st;
 175         int saved_errno = 0;
 176         uint64_t hash_test;
 177         unsigned v;
 178         ssize_t rlen;
 179         struct tdb_header hdr;
 180         struct tdb_attribute_seed *seed = NULL;
 181
 182         tdb = malloc(sizeof(*tdb));
 183         if (!tdb) {
 184                 /* Can't log this */
 185                 errno = ENOMEM;
 186                 return NULL;
 187         }
 188         tdb->name = NULL;
 189         tdb->map_ptr = NULL;
 190         tdb->direct_access = 0;
 191         tdb->fd = -1;
 192         tdb->map_size = sizeof(struct tdb_header);
 193         tdb->ecode = TDB_SUCCESS;
 194         tdb->flags = tdb_flags;
 195         tdb->logfn = NULL;
 196         tdb->transaction = NULL;
 197         tdb->stats = NULL;
 198         tdb->access = NULL;
 199         tdb_hash_init(tdb);
 200         tdb_io_init(tdb);
 201         tdb_lock_init(tdb);
 202
 203         while (attr) {
 204                 switch (attr->base.attr) {
 205                 case TDB_ATTRIBUTE_LOG:
 206                         tdb->logfn = attr->log.log_fn;
 207                         tdb->log_private = attr->log.log_private;
 208                         break;
 209                 case TDB_ATTRIBUTE_HASH:
 210                         tdb->khash = attr->hash.hash_fn;
 211                         tdb->hash_priv = attr->hash.hash_private;
 212                         break;
 213                 case TDB_ATTRIBUTE_SEED:
 214                         seed = &attr->seed;
 215                         break;
 216                 case TDB_ATTRIBUTE_STATS:
 217                         tdb->stats = &attr->stats;
 218                         /* They have stats we don't know about?  Tell them. */
 219                         if (tdb->stats->size > sizeof(attr->stats))
 220                                 tdb->stats->size = sizeof(attr->stats);
 221                         break;
 222                 default:
 223                         tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
 224                                    "tdb_open: unknown attribute type %u",
 225                                    attr->base.attr);
 226                         goto fail;
 227                 }
 228                 attr = attr->base.next;
 229         }
 230
 231         if ((open_flags & O_ACCMODE) == O_WRONLY) {
 232                 tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
 233                            "tdb_open: can't open tdb %s write-only", name);
 234                 goto fail;
 235         }
 236
 237         if ((open_flags & O_ACCMODE) == O_RDONLY) {
 238                 tdb->read_only = true;
 239                 /* read only databases don't do locking */
 240                 tdb->flags |= TDB_NOLOCK;
 241                 tdb->mmap_flags = PROT_READ;
 242         } else {
 243                 tdb->read_only = false;
 244                 tdb->mmap_flags = PROT_READ | PROT_WRITE;
 245         }
 246
 247         /* internal databases don't need any of the rest. */
 248         if (tdb->flags & TDB_INTERNAL) {
 249                 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
 250                 if (tdb_new_database(tdb, seed, &hdr) != 0) {
 251                         goto fail;
 252                 }
 253                 tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
 254                 tdb->hash_seed = hdr.hash_seed;
 255                 tdb_ftable_init(tdb);
 256                 return tdb;
 257         }
 258
 259         if ((tdb->fd = open(name, open_flags, mode)) == -1) {
 260                 /* errno set by open(2) */
 261                 saved_errno = errno;
 262                 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 263                            "tdb_open: could not open file %s: %s",
 264                            name, strerror(errno));
 265                 goto fail;
 266         }
 267
 268         /* on exec, don't inherit the fd */
 269         v = fcntl(tdb->fd, F_GETFD, 0);
 270         fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
 271
 272         /* ensure there is only one process initialising at once */
 273         if (tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK) == -1) {
 274                 /* errno set by tdb_brlock */
 275                 saved_errno = errno;
 276                 goto fail;
 277         }
 278
 279         /* If they used O_TRUNC, read will return 0. */
 280         rlen = read(tdb->fd, &hdr, sizeof(hdr));
 281         if (rlen == 0 && (open_flags & O_CREAT)) {
 282                 if (tdb_new_database(tdb, seed, &hdr) == -1) {
 283                         goto fail;
 284                 }
 285         } else if (rlen < 0) {
 286                 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 287                            "tdb_open: error %s reading %s",
 288                            strerror(errno), name);
 289                 goto fail;
 290         } else if (rlen < sizeof(hdr)
 291                    || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
 292                 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 293                            "tdb_open: %s is not a tdb file", name);
 294                 goto fail;
 295         }
 296
 297         if (hdr.version != TDB_VERSION) {
 298                 if (hdr.version == bswap_64(TDB_VERSION))
 299                         tdb->flags |= TDB_CONVERT;
 300                 else {
 301                         /* wrong version */
 302                         tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 303                                    "tdb_open: %s is unknown version 0x%llx",
 304                                    name, (long long)hdr.version);
 305                         goto fail;
 306                 }
 307         }
 308
 309         tdb_convert(tdb, &hdr, sizeof(hdr));
 310         tdb->hash_seed = hdr.hash_seed;
 311         hash_test = TDB_HASH_MAGIC;
 312         hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
 313         if (hdr.hash_test != hash_test) {
 314                 /* wrong hash variant */
 315                 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 316                            "tdb_open: %s uses a different hash function",
 317                            name);
 318                 goto fail;
 319         }
 320
 321         if (fstat(tdb->fd, &st) == -1) {
 322                 saved_errno = errno;
 323                 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 324                            "tdb_open: could not stat open %s: %s",
 325                            name, strerror(errno));
 326                 goto fail;
 327         }
 328
 329         /* Is it already in the open list?  If so, fail. */
 330         if (tdb_already_open(st.st_dev, st.st_ino)) {
 331                 /* FIXME */
 332                 tdb_logerr(tdb, TDB_ERR_NESTING, TDB_DEBUG_ERROR,
 333                            "tdb_open: %s (%d,%d) is already open in this"
 334                            " process",
 335                            name, (int)st.st_dev, (int)st.st_ino);
 336                 goto fail;
 337         }
 338
 339         tdb->name = strdup(name);
 340         if (!tdb->name) {
 341                 tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_ERROR,
 342                            "tdb_open: failed to allocate name");
 343                 goto fail;
 344         }
 345
 346         tdb->device = st.st_dev;
 347         tdb->inode = st.st_ino;
 348         tdb_unlock_open(tdb);
 349
 350         /* This make sure we have current map_size and mmap. */
 351         tdb->methods->oob(tdb, tdb->map_size + 1, true);
 352
 353         /* Now it's fully formed, recover if necessary. */
 354         if (tdb_needs_recovery(tdb) && tdb_lock_and_recover(tdb) == -1) {
 355                 goto fail;
 356         }
 357
 358         if (tdb_ftable_init(tdb) == -1)
 359                 goto fail;
 360
 361         tdb->next = tdbs;
 362         tdbs = tdb;
 363         return tdb;
 364
 365  fail:
 366         /* Map ecode to some logical errno. */
 367         if (!saved_errno) {
 368                 switch (tdb->ecode) {
 369                 case TDB_ERR_CORRUPT:
 370                 case TDB_ERR_IO:
 371                         saved_errno = EIO;
 372                         break;
 373                 case TDB_ERR_LOCK:
 374                         saved_errno = EWOULDBLOCK;
 375                         break;
 376                 case TDB_ERR_OOM:
 377                         saved_errno = ENOMEM;
 378                         break;
 379                 case TDB_ERR_EINVAL:
 380                         saved_errno = EINVAL;
 381                         break;
 382                 case TDB_ERR_NESTING:
 383                         saved_errno = EBUSY;
 384                         break;
 385                 default:
 386                         saved_errno = EINVAL;
 387                         break;
 388                 }
 389         }
 390
 391 #ifdef TDB_TRACE
 392         close(tdb->tracefd);
 393 #endif
 394         if (tdb->map_ptr) {
 395                 if (tdb->flags & TDB_INTERNAL) {
 396                         free(tdb->map_ptr);
 397                 } else
 398                         tdb_munmap(tdb);
 399         }
 400         free(tdb->lockrecs);
 401         free((char *)tdb->name);
 402         if (tdb->fd != -1)
 403                 if (close(tdb->fd) != 0)
 404                         tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 405                                    "tdb_open: failed to close tdb->fd"
 406                                    " on error!");
 407         free(tdb);
 408         errno = saved_errno;
 409         return NULL;
 410 }
 411
 412 static int update_rec_hdr(struct tdb_context *tdb,
 413                           tdb_off_t off,
 414                           tdb_len_t keylen,
 415                           tdb_len_t datalen,
 416                           struct tdb_used_record *rec,
 417                           uint64_t h)
 418 {
 419         uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
 420
 421         if (set_header(tdb, rec, TDB_USED_MAGIC, keylen, datalen,
 422                        keylen + dataroom, h))
 423                 return -1;
 424
 425         return tdb_write_convert(tdb, off, rec, sizeof(*rec));
 426 }
 427
 428 /* Returns -1 on error, 0 on OK */
 429 static int replace_data(struct tdb_context *tdb,
 430                         struct hash_info *h,
 431                         struct tdb_data key, struct tdb_data dbuf,
 432                         tdb_off_t old_off, tdb_len_t old_room,
 433                         bool growing)
 434 {
 435         tdb_off_t new_off;
 436
 437         /* Allocate a new record. */
 438         new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, TDB_USED_MAGIC,
 439                         growing);
 440         if (unlikely(new_off == TDB_OFF_ERR))
 441                 return -1;
 442
 443         /* We didn't like the existing one: remove it. */
 444         if (old_off) {
 445                 add_stat(tdb, frees, 1);
 446                 add_free_record(tdb, old_off,
 447                                 sizeof(struct tdb_used_record)
 448                                 + key.dsize + old_room);
 449                 if (replace_in_hash(tdb, h, new_off) == -1)
 450                         return -1;
 451         } else {
 452                 if (add_to_hash(tdb, h, new_off) == -1)
 453                         return -1;
 454         }
 455
 456         new_off += sizeof(struct tdb_used_record);
 457         if (tdb->methods->twrite(tdb, new_off, key.dptr, key.dsize) == -1)
 458                 return -1;
 459
 460         new_off += key.dsize;
 461         if (tdb->methods->twrite(tdb, new_off, dbuf.dptr, dbuf.dsize) == -1)
 462                 return -1;
 463
 464         /* FIXME: tdb_increment_seqnum(tdb); */
 465         return 0;
 466 }
 467
 468 int tdb_store(struct tdb_context *tdb,
 469               struct tdb_data key, struct tdb_data dbuf, int flag)
 470 {
 471         struct hash_info h;
 472         tdb_off_t off;
 473         tdb_len_t old_room = 0;
 474         struct tdb_used_record rec;
 475         int ret;
 476
 477         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 478         if (unlikely(off == TDB_OFF_ERR))
 479                 return -1;
 480
 481         /* Now we have lock on this hash bucket. */
 482         if (flag == TDB_INSERT) {
 483                 if (off) {
 484                         tdb->ecode = TDB_ERR_EXISTS;
 485                         goto fail;
 486                 }
 487         } else {
 488                 if (off) {
 489                         old_room = rec_data_length(&rec)
 490                                 + rec_extra_padding(&rec);
 491                         if (old_room >= dbuf.dsize) {
 492                                 /* Can modify in-place.  Easy! */
 493                                 if (update_rec_hdr(tdb, off,
 494                                                    key.dsize, dbuf.dsize,
 495                                                    &rec, h.h))
 496                                         goto fail;
 497                                 if (tdb->methods->twrite(tdb, off + sizeof(rec)
 498                                                          + key.dsize,
 499                                                          dbuf.dptr, dbuf.dsize))
 500                                         goto fail;
 501                                 tdb_unlock_hashes(tdb, h.hlock_start,
 502                                                   h.hlock_range, F_WRLCK);
 503                                 return 0;
 504                         }
 505                 } else {
 506                         if (flag == TDB_MODIFY) {
 507                                 /* if the record doesn't exist and we
 508                                    are in TDB_MODIFY mode then we should fail
 509                                    the store */
 510                                 tdb->ecode = TDB_ERR_NOEXIST;
 511                                 goto fail;
 512                         }
 513                 }
 514         }
 515
 516         /* If we didn't use the old record, this implies we're growing. */
 517         ret = replace_data(tdb, &h, key, dbuf, off, old_room, off != 0);
 518         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 519         return ret;
 520
 521 fail:
 522         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 523         return -1;
 524 }
 525
 526 int tdb_append(struct tdb_context *tdb,
 527                struct tdb_data key, struct tdb_data dbuf)
 528 {
 529         struct hash_info h;
 530         tdb_off_t off;
 531         struct tdb_used_record rec;
 532         tdb_len_t old_room = 0, old_dlen;
 533         unsigned char *newdata;
 534         struct tdb_data new_dbuf;
 535         int ret;
 536
 537         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 538         if (unlikely(off == TDB_OFF_ERR))
 539                 return -1;
 540
 541         if (off) {
 542                 old_dlen = rec_data_length(&rec);
 543                 old_room = old_dlen + rec_extra_padding(&rec);
 544
 545                 /* Fast path: can append in place. */
 546                 if (rec_extra_padding(&rec) >= dbuf.dsize) {
 547                         if (update_rec_hdr(tdb, off, key.dsize,
 548                                            old_dlen + dbuf.dsize, &rec, h.h))
 549                                 goto fail;
 550
 551                         off += sizeof(rec) + key.dsize + old_dlen;
 552                         if (tdb->methods->twrite(tdb, off, dbuf.dptr,
 553                                                  dbuf.dsize) == -1)
 554                                 goto fail;
 555
 556                         /* FIXME: tdb_increment_seqnum(tdb); */
 557                         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range,
 558                                           F_WRLCK);
 559                         return 0;
 560                 }
 561
 562                 /* Slow path. */
 563                 newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
 564                 if (!newdata) {
 565                         tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
 566                                    "tdb_append: failed to allocate %zu bytes",
 567                                    (size_t)(key.dsize+old_dlen+dbuf.dsize));
 568                         goto fail;
 569                 }
 570                 if (tdb->methods->tread(tdb, off + sizeof(rec) + key.dsize,
 571                                         newdata, old_dlen) != 0) {
 572                         free(newdata);
 573                         goto fail;
 574                 }
 575                 memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
 576                 new_dbuf.dptr = newdata;
 577                 new_dbuf.dsize = old_dlen + dbuf.dsize;
 578         } else {
 579                 newdata = NULL;
 580                 new_dbuf = dbuf;
 581         }
 582
 583         /* If they're using tdb_append(), it implies they're growing record. */
 584         ret = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
 585         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 586         free(newdata);
 587
 588         return ret;
 589
 590 fail:
 591         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 592         return -1;
 593 }
 594
 595 struct tdb_data tdb_fetch(struct tdb_context *tdb, struct tdb_data key)
 596 {
 597         tdb_off_t off;
 598         struct tdb_used_record rec;
 599         struct hash_info h;
 600         struct tdb_data ret;
 601
 602         off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
 603         if (unlikely(off == TDB_OFF_ERR))
 604                 return tdb_null;
 605
 606         if (!off) {
 607                 tdb->ecode = TDB_ERR_NOEXIST;
 608                 ret = tdb_null;
 609         } else {
 610                 ret.dsize = rec_data_length(&rec);
 611                 ret.dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
 612                                           ret.dsize);
 613         }
 614
 615         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
 616         return ret;
 617 }
 618
 619 int tdb_delete(struct tdb_context *tdb, struct tdb_data key)
 620 {
 621         tdb_off_t off;
 622         struct tdb_used_record rec;
 623         struct hash_info h;
 624
 625         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 626         if (unlikely(off == TDB_OFF_ERR))
 627                 return -1;
 628
 629         if (!off) {
 630                 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 631                 tdb->ecode = TDB_ERR_NOEXIST;
 632                 return -1;
 633         }
 634
 635         if (delete_from_hash(tdb, &h) == -1)
 636                 goto unlock_err;
 637
 638         /* Free the deleted entry. */
 639         add_stat(tdb, frees, 1);
 640         if (add_free_record(tdb, off,
 641                             sizeof(struct tdb_used_record)
 642                             + rec_key_length(&rec)
 643                             + rec_data_length(&rec)
 644                             + rec_extra_padding(&rec)) != 0)
 645                 goto unlock_err;
 646
 647         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 648         return 0;
 649
 650 unlock_err:
 651         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 652         return -1;
 653 }
 654
 655 int tdb_close(struct tdb_context *tdb)
 656 {
 657         struct tdb_context **i;
 658         int ret = 0;
 659
 660         tdb_trace(tdb, "tdb_close");
 661
 662         if (tdb->transaction) {
 663                 tdb_transaction_cancel(tdb);
 664         }
 665
 666         if (tdb->map_ptr) {
 667                 if (tdb->flags & TDB_INTERNAL)
 668                         free(tdb->map_ptr);
 669                 else
 670                         tdb_munmap(tdb);
 671         }
 672         free((char *)tdb->name);
 673         if (tdb->fd != -1) {
 674                 ret = close(tdb->fd);
 675                 tdb->fd = -1;
 676         }
 677         free(tdb->lockrecs);
 678
 679         /* Remove from contexts list */
 680         for (i = &tdbs; *i; i = &(*i)->next) {
 681                 if (*i == tdb) {
 682                         *i = tdb->next;
 683                         break;
 684                 }
 685         }
 686
 687 #ifdef TDB_TRACE
 688         close(tdb->tracefd);
 689 #endif
 690         free(tdb);
 691
 692         return ret;
 693 }
 694
 695 enum TDB_ERROR tdb_error(const struct tdb_context *tdb)
 696 {
 697         return tdb->ecode;
 698 }
 699
 700 const char *tdb_errorstr(const struct tdb_context *tdb)
 701 {
 702         /* Gcc warns if you miss a case in the switch, so use that. */
 703         switch (tdb->ecode) {
 704         case TDB_SUCCESS: return "Success";
 705         case TDB_ERR_CORRUPT: return "Corrupt database";
 706         case TDB_ERR_IO: return "IO Error";
 707         case TDB_ERR_LOCK: return "Locking error";
 708         case TDB_ERR_OOM: return "Out of memory";
 709         case TDB_ERR_EXISTS: return "Record exists";
 710         case TDB_ERR_NESTING: return "Transaction already started";
 711         case TDB_ERR_EINVAL: return "Invalid parameter";
 712         case TDB_ERR_NOEXIST: return "Record does not exist";
 713         case TDB_ERR_RDONLY: return "write not permitted";
 714         }
 715         return "Invalid error code";
 716 }
 717
 718 void COLD tdb_logerr(struct tdb_context *tdb,
 719                      enum TDB_ERROR ecode,
 720                      enum tdb_debug_level level,
 721                      const char *fmt, ...)
 722 {
 723         char *message;
 724         va_list ap;
 725         size_t len;
 726         /* tdb_open paths care about errno, so save it. */
 727         int saved_errno = errno;
 728
 729         tdb->ecode = ecode;
 730
 731         if (!tdb->logfn)
 732                 return;
 733
 734         /* FIXME: Doesn't assume asprintf. */
 735         va_start(ap, fmt);
 736         len = vsnprintf(NULL, 0, fmt, ap);
 737         va_end(ap);
 738
 739         message = malloc(len + 1);
 740         if (!message) {
 741                 tdb->logfn(tdb, TDB_DEBUG_ERROR, tdb->log_private,
 742                            "out of memory formatting message:");
 743                 tdb->logfn(tdb, level, tdb->log_private, fmt);
 744                 return;
 745         }
 746         va_start(ap, fmt);
 747         len = vsprintf(message, fmt, ap);
 748         va_end(ap);
 749         tdb->logfn(tdb, level, tdb->log_private, message);
 750         free(message);
 751         errno = saved_errno;
 752 }